Skip to content

phuselab/pyVHR

Repository files navigation

pyVHR logo

PyPI - Python Version PyPI GitHub last commit GitHub license

Package pyVHR (short for Python framework for Virtual Heart Rate) is a comprehensive framework for studying methods of pulse rate estimation relying on video, also known as remote photoplethysmography (rPPG).

pyVHR: a Python framework for remote photoplethysmography

Description

The methodological rationale behind the framework is that in order to study, develop and compare new rPPG methods in a principled and reproducible way, the following conditions should be met: i) a structured pipeline to monitor rPPG algorithms' input, output, and main control parameters; ii) the availability and the use of multiple datasets; iii) a sound statistical assessment of methods' performance. pyVHR allows to easily handle rPPGmethods and data, while simplifying the statistical assessment. Its main features lie in the following:

  • Analysis-oriented. It constitutes a platform for experiment design, involving an arbitrary number of methods applied to multiple video datasets. It provides a systemic end-to-end pipeline, allowing to assess different rPPG algorithms, by easily setting parameters and meta-parameters.
  • Openness. It comprises both method and dataset factory, so to easily extend the pool of elements to be evaluatedwith newly developed rPPG methods and any kind of videodatasets.
  • Robust assessment. The outcomes are arranged intostructured data ready for in-depth analyses. Performance comparison is carried out based on robust nonparametric statistical tests.

Nine classical rPPG methods, namely ICA, PCA, GREEN, CHROM, POS, SSR, LGI, PBV, OMIT, as well as the recent Deep Learning-based model MTTS-CAN are implemented. Moreover, pyVHR provides APIs for handling 11 publicly available video datasets, i.e. PURE, LGI-PPGI-DB, UBFC-1, , UBFC-2, UBFC-Phys, ECG-Fitness, MAHNOB Vicar-PPG-2, V4V , VIPL-HR and COHFACE, usually adopted to benchmark rPPG methods. Eventually, extensive rigorous statistical analyses can be effortlessly performed via the pyVHR stats APIs.

pipeline

Getting started

Dependencies

The quickest way to get started is to install the miniconda distribution, a lightweight minimal installation of Anaconda Python.

Once installed, create a new conda environment and automatically fetch all the dependencies based on your architecture (with or without GPU), using one of the following commands:

CPU-only version (v. 1.2 - previous version)

conda env create --file https://raw.githubusercontent.com/phuselab/pyVHR/pyVHR_CPU/pyVHR_CPU_env.yml

CPU+GPU version (v. 2.0 - current version)

This yml environment is for cudatoolkit=11.3 and python=3.9.

conda env create --file https://raw.githubusercontent.com/phuselab/pyVHR/master/pyVHR_env.yml

Installation

Enter the newly created conda environment and install the latest stable release build of pyVHR with:

CPU-only version (v. 1.2 - previous version)

conda activate pyvhr
(pyvhr) pip install pyvhr-cpu

CPU+GPU version (v. 2.0 - current version)

conda activate pyvhr
(pyvhr) pip install pyvhr

Basic usage

Run the following code to obtain BPM estimates over time for a single video:

from pyVHR.analysis.pipeline import Pipeline
from pyVHR.plot.visualize import *
from pyVHR.utils.errors import getErrors, printErrors, displayErrors

# params
wsize = 6                  # window size in seconds
roi_approach = 'patches'   # 'holistic' or 'patches'
bpm_est = 'clustering'     # BPM final estimate, if patches choose 'medians' or 'clustering'
method = 'cpu_CHROM'       # one of the methods implemented in pyVHR

# run
pipe = Pipeline()          # object to execute the pipeline
bvps, timesES, bpmES = pipe.run_on_video(videoFileName,
                                        winsize=wsize, 
                                        roi_method='convexhull',
                                        roi_approach=roi_approach,
                                        method=method,
                                        estimate=bpm_est,
                                        patch_size=0, 
                                        RGB_LOW_HIGH_TH=(5,230),
                                        Skin_LOW_HIGH_TH=(5,230),
                                        pre_filt=True,
                                        post_filt=True,
                                        cuda=True, 
                                        verb=True)

# ERRORS
RMSE, MAE, MAX, PCC, CCC, SNR = getErrors(bvps, fps, bpmES, bpmGT, timesES, timesGT)
printErrors(RMSE, MAE, MAX, PCC, CCC, SNR)
displayErrors(bpmES, bpmGT, timesES, timesGT)

The full documentation of run_on_video method, with all the possible parameters, can be found here: https://phuselab.github.io/pyVHR/

Notebooks

Some demonstration jupyter notebooks that help to better understand the many features of the framework are contained in the notebooks folder.

  • pyVHR_demo.ipynb: Basic demo with individual steps explained in detail.
  • pyVHR_run_on_video.ipynb: Show execution on a single video by deriving HRVs and error values from the reference signal.
  • pyVHR_run_on_dataset.ipynb: Show execution on a single dataset by deriving HRVs and error values from the reference signals. It is also possible to make some basic statistics, boxplots and ranking tests for comparative purposes.
  • pyVHR_demo_deep.ipynb: Show execution of deep methods on a single dataset by deriving HRV and error values from reference signals.

Documentation

The full documentation of the pyVHR framework is available at https://phuselab.github.io/pyVHR/.

Developing

The latest unstable development build of pyVHR is available on GitHub, and can be obtained downloading from source and installing via:

git clone [email protected]:phuselab/pyVHR.git
cd pyVHR/
python setup.py install

The main branch refers to the full pyVHR framework (requires GPU), while the pyVHR_CPU branch is dedicated to the CPU-only architectures.

Custom installation

If you want to create your environment from scratch you should follow these steps:

  • Install PyTorch (here)
  • Install Numba (here)
  • Install Cupy (for GPU only) with the correct CUDA version (here)
  • Install CuSignal (for GPU only) using conda and remove from the command 'cudatoolkit=x.y' (here)
  • Install Kaleido (here)
  • Install PyTables (here)
  • Install pyVHR as shown above.

Methods

The framework contains the implementation of many common methods for remote-PPG measurement. Currently implemented methods with reference publications are:

Method name Reference paper
Green Verkruysse, W., Svaasand, L. O., & Nelson, J. S. (2008). Remote plethysmographic imaging using ambient light. Optics express, 16(26), 21434-21445.
CHROM De Haan, G., & Jeanne, V. (2013). Robust pulse rate from chrominance-based rPPG. IEEE Transactions on Biomedical Engineering, 60(10), 2878-2886.
ICA Poh, M. Z., McDuff, D. J., & Picard, R. W. (2010). Non-contact, automated cardiac pulse measurements using video imaging and blind source separation. Optics express, 18(10), 10762-10774.
LGI Pilz, C. S., Zaunseder, S., Krajewski, J., & Blazek, V. (2018). Local group invariance for heart rate estimation from face videos in the wild. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (pp. 1254-1262).
PBV De Haan, G., & Van Leest, A. (2014). Improved motion robustness of remote-PPG by using the blood volume pulse signature. Physiological measurement, 35(9), 1913.
PCA Lewandowska, M., Rumiński, J., Kocejko, T., & Nowak, J. (2011, September). Measuring pulse rate with a webcam—a non-contact method for evaluating cardiac activity. In 2011 federated conference on computer science and information systems (FedCSIS) (pp. 405-410). IEEE.
POS Wang, W., den Brinker, A. C., Stuijk, S., & de Haan, G. (2016). Algorithmic principles of remote PPG. IEEE Transactions on Biomedical Engineering, 64(7), 1479-1491.
SSR Wang, W., Stuijk, S., & De Haan, G. (2015). A novel algorithm for remote photoplethysmography: Spatial subspace rotation. IEEE transactions on biomedical engineering, 63(9), 1974-1984.
OMIT Álvarez Casado, C., Bordallo López, M. (2022). Face2PPG: An unsupervised pipeline for blood volume pulse extraction from faces. arXiv (eprint 2202.04101).
MTTS-CAN Liu, X., Fromm, J., Patel, S., & McDuff, D. (2020). Multi-task temporal shift attention networks for on-device contactless vitals measurement. Advances in Neural Information Processing Systems, 33, 19400-19411.
HR-CNN Spetlik, R., Franc, V., Cech, J. and Matas, J. (2018). Visual Heart Rate Estimation with Convolutional Neural Network. In Proceedings of British Machine Vision Conference

Datasets

Interfaces for 10 different datasets are provided in the datasets folder. Once the datasets are obtained, the respective files must be edited to match the correct path.
Currently supported datasets are:

Dataset name Link
COHFACE https://www.idiap.ch/dataset/cohface
LGI-PPGI https://github.com/partofthestars/LGI-PPGI-DB
MAHNOB-HCI https://mahnob-db.eu/hci-tagging/
PURE https://www.tu-ilmenau.de/neurob/data-sets-code/pulse-rate-detection-dataset-pure
UBFC1 https://sites.google.com/view/ybenezeth/ubfcrppg
UBFC2 https://sites.google.com/view/ybenezeth/ubfcrppg
UBFC-Phys https://sites.google.com/view/ybenezeth/ubfc-phys
ECG-Fitness https://cmp.felk.cvut.cz/~spetlrad/ecg-fitness/*
Vicar-PPG-2 https://docs.google.com/forms/d/e/1FAIpQLScwnW_D5M4JVovPzpxA0Bf1ZCTaG5vh7sYu48I0MVSpgltvdw/viewform*
V4V https://vision4vitals.github.io/
VIPL-HR https://arxiv.org/abs/1810.04927

RESULTS

Here are the results obtained (holistic vs median vs clustering) by applying the pyVHR_run_on_dataset notebook to some datasets listed above:

Dataset MAE Error PCC Error
PURE PURE_MAE PURE_PCC
UBFC1 UBFC1_MAE UBFC1_PCC
UBFC2 UBFC2_MAE UBFC2_PCC
LGI-PPGI LGI-PPGI_MAE LGI-PPGI_PCC
ECG_Fitness_01-1 ECG_Fitness_01-1_MAE ECG_Fitness_01-1_PCC
ECG_Fitness_01-2 ECG_Fitness_01-2_MAE ECG_Fitness_01-2_PCC
ECG_Fitness_02-1 ECG_Fitness_02-1_MAE ECG_Fitness_02-1_PCC
ECG_Fitness_02-2 ECG_Fitness_02-2_MAE ECG_Fitness_02-2_PCC
ECG_Fitness_03-1 ECG_Fitness_03-1_MAE ECG_Fitness_03-1_PCC
ECG_Fitness_03-2 ECG_Fitness_03-2_MAE ECG_Fitness_03-2_PCC
ECG_Fitness_04-1 ECG_Fitness_04-1_MAE ECG_Fitness_04-1_PCC
ECG_Fitness_04-2 ECG_Fitness_04-2_MAE ECG_Fitness_04-2_PCC
ECG_Fitness_05-1 ECG_Fitness_05-1_MAE ECG_Fitness_05-1_PCC
ECG_Fitness_05-2 ECG_Fitness_05-2_MAE ECG_Fitness_05-2_PCC
ECG_Fitness_06-1 ECG_Fitness_06-1_MAE ECG_Fitness_06-1_PCC
ECG_Fitness_06-2 ECG_Fitness_06-2_MAE ECG_Fitness_06-2_PCC

GUI

In the folder realtime you can find an example of a simple GUI created using the pyVHR package. You can launch it by going into the path pyVHR/realtime/ and using the command

python GUI.py

If you want to use a specific rPPG method and pre-post filterings, you must set them in the last lines of GUI.py.

Below is a video showing the use of the GUI.

GUI.mp4

Reference

If you use this code, please cite the papers:

@article{boccignone2022pyvhr,
  title={pyVHR: a Python framework for remote photoplethysmography},
  author={Boccignone, Giuseppe and Conte, Donatello and Cuculo, Vittorio and D’Amelio, Alessandro and Grossi, Giuliano and Lanzarotti, Raffaella and Mortara, Edoardo},
  journal={PeerJ Computer Science},
  volume={8},
  pages={e929},
  year={2022},
  publisher={PeerJ Inc.}
}
@article{Boccignone2020,
  doi = {10.1109/access.2020.3040936},
  url = {https://doi.org/10.1109/access.2020.3040936},
  year = {2020},
  publisher = {Institute of Electrical and Electronics Engineers ({IEEE})},
  pages = {1--1},
  author = {Giuseppe Boccignone and Donatello Conte and Vittorio Cuculo and Alessandro D’Amelio and Giuliano Grossi and Raffaella Lanzarotti},
  title = {An Open Framework for Remote-{PPG} Methods and their Assessment},
  journal = {{IEEE} Access}
}

License

This project is licensed under the GPL-3.0 License - see the LICENSE file for details